Structural controls on the hydrogeological functioning of a floodplain

Sedimentary cores were collected both with a sonic drilling unit (SmallRotoSonic SRS T, SonicSampDrill) and a direct push unit (6610 DT, Geoprobe). The depth of the drill cores was limited either by lack of penetration into competent bedrock, or legal regulations from drilling permits. The collected drill cores (51 mm in diameter) were stored in plastic liners or metal core boxes. Preparation of the cores involved exposing fresh surfaces either by hand, or by freezing the core and subsequently cutting it with a rock saw. The internal structure was described, photographed and grouped into main lithologies. For selected cores, a grain-size analysis was performed. As per German Standard DIN 18123-4, sieve analysis was conducted for the grain sizes greater than 0.063 mm (coarse fraction), with sedimentation analysis being conducted for the clay- and silt-sized grains (fine fraction). Samples of 100–200 g were collected in fixed 50-cm intervals to prevent sampling bias in the vertical profiles of grain-size distribution.

A 550-m long Wenner-α electrical resistivity profile was acquired with 1.5-m electrode spacing across the floodplain (RESECS, DMT-Group). The data were subsequently filtered and inverted with PyBERT to receive a continuous profile of subsurface resistivity distribution and infer layer continuity and boundaries for the geological model (Günther et al. 2006). Based on the inversion result, a spatial geoelectrical mapping of a higher resistive target feature with fixed electrode spacings was conducted, as described by Klingler et al. 2020. Vertical geophysical logging was also performed, including direct-push electrical-conductivity (EC) profiles, direct-push color logging, and logging of natural gamma radiation in boreholes. The logging results mainly confirmed the major stratification of Quaternary floodplain sediments discussed in the following.

The geological mapping incorporated hard data from lithofacies picks (both drill core data and vertical gamma profiles), along with regional geological datasets, into the implicit geological modelling software LeapFrog (LeapFrog Works). The depths of the facies contacts were converted to elevations for the purposes of correlation. The boundaries of the geological model, i.e. contact between floodplain sediments and subcropping bedrock units, were defined using the most recent 1:50,000 regional geologic map (LGRB 2012). The model top is taken from topography of the study area by means of a 10-m digital elevation model (LGL 2009). The base of the geological model is the subcropping bedrock, where minimal hard data is available, necessitating the use of polylines within LeapFrog, mimicking typical bedrock valley shapes, to guide the bedrock surface interpretation. LeapFrog invokes implicit modelling in which geological contact surfaces are created by interpolation of observed lithofacies contacts and their orientation. With the ground surface and bedrock topography as the upper and lower boundary of the model, respectively, volumes of the geological units are generated between the sedimentary surfaces, and cross-sections can be generated for any region of the modelling domain.

Monitoring wells were installed by advancing an 83 mm (3.25″) metal casing (with an expandable tip) to the target depth using a direct push rig, then lowering the assembled monitoring well string before pulling back the metal casing. Monitoring-well materials at the site include 51 mm (2″) diameter polyvinyl chloride, as well as 25 mm (1″) and 41 mm (1.5″) nominal diameter high-density polyethylene well casings. Well screens were completed with a poured sand pack, prepacked sand pack, or by natural formation collapse, which was always followed by a bentonite top seal prior to backfilling the annular space. A flush-mount protective casing was installed at all monitoring wells, with a cement and concrete seal to prevent surface-water infiltration. Monitoring wells were developed using suction pumping and surging methods to improve hydraulic communication to the surrounding sediments disturbed during the process of well installation. The top of the well casing and ground surface were surveyed using a combination of a differential global positioning system and a laser level.

Figure 2 shows the locations of a weather station in Unterjesingen (LTZ 2020) and a gauging station for Ammer River at Pfäffingen (LUBW 2020). Groundwater level data in monitoring wells were continuously collected with dedicated groundwater data loggers (CTD and MicroDiver, Schlumberger Water Services) in 15-min intervals; or, when pressure loggers were not available, measured with a manual water level tape. The measured absolute pressure values were corrected using the barometric pressure from a Baro-Diver approximately 7 km from the site, in Tübingen. Each pressure-logger dataset was manually reviewed, corrected for obvious errors, and converted to absolute groundwater elevations with the manual water-level readings. Groundwater flow maps were constructed by spatial interpolation of groundwater elevations for a specific timestamp, using ordinary kriging with an omnidirectional linear variogram.

Hydraulic parameters in the vicinity of the monitoring wells were estimated by well testing, including slug tests, drawdown-recovery tests, and multi-well pumping tests, each sampling a different volume of the aquifer. Analysis of the slug and pumping test consisted of type-curve regression fits on the measured displacement time-series with the appropriate analytical solution (HydroSOLVE Inc.): slug tests showing an overdampened water-level response were analyzed using the Bouwer and Rice (1976) or Hyder et al. (1994) analytical solutions, whereas tests with an underdampened (oscillatory) response required the Butler Jr. (1998) analytical solution to fit the data. Late-time recovery data from single-well recovery tests were fit with the Theis Recovery (Theis 1935) method, and drawdown and recovery data from multi-well pumping tests were fit with the Theis (1935) analytical solution.

The monitoring wells were periodically sampled throughout the monitoring period of the study. With a battery-powered peristaltic pump (Eijkelkamp Soil & Water) all sampled monitoring wells were purged prior to collecting a representative groundwater sample. For all groundwater samples, field-measured parameters were either recorded continuously with a flow-through cell, instrumented with a multi-parameter probe (smarTROLL and Aqua TROLL 500, In-Situ), or from the recovering groundwater level after purging the monitoring well dry. Field-measured parameters included pH, temperature, specific electrical conductivity (EC), dissolved oxygen (DO), and redox potential. Same-day filtered groundwater samples were analyzed for major ion hydrochemistry with ion chromatography (DX-120, Dionex), dissolved organic carbon (HighTOC, Elementar), alkalinity (Titrino Plus, Metrohm), and nitrogen species (AutoAnalyzer 3 HR, SEAL Analytical). Due to the instability of ammonium, the nitrogen species were analyzed on the next morning. Bisulfide was analyzed separately (Multiskan GO, Thermo Scientific) for a single sampling event in July 2019, where the samples were stabilized in the field using a zinc acetate solution and frozen as soon as possible. Erroneous results from sampling campaigns were flagged during the review of the hydrochemistry datasets. This process included checking the ion balance and a comparison to the previous hydrochemical datasets. Statistical analysis of the hydrochemical data included reviewing univariate and bivariate distributions, as well as a hierarchical cluster analysis (HCA) using the Euclidean distance (Ward 1963). Prior to the cluster analysis, the cleaned hydrochemical datasets typically required log-transformation, as well as a z-score normalization. As with the groundwater contour map, interpretation of spatial hydrochemical patterns involved interpolation using ordinary kriging, selecting a variogram model to best fit the spatial correlation structure of the hydrochemical parameter.

Figure 3 presents the central floodplain lithofacies as a standard lithostratigraphic profile including grain size and total organic carbon analysis along with representative photos of drill cores. The average depth to bedrock is 14 m, which is retrieved as a reddish-grayish laminated mudstone in a few cores. Drilling permits and equipment limited the penetration into bedrock, and thus, when there was no bedrock recovery, the lithologic contact was assumed at the maximum drilling depth.

“Gravel” is the lowermost Quaternary lithofacies, which was recovered typically from 11 to 14 m below ground surface, and consists of poorly sorted clasts up to small cobble size within a gray, clayey-silty matrix (Fig. 3d). The coarse fraction is dominated by well-rounded limestone clasts, as well as a few sandstone and mudstone clasts. As the bedrock at the hillslopes contains no limestones, these clasts must have originated from the Middle Triassic Muschelkalk limestone outcrops upstream of Pfäffingen (Fig. 2) and been deposited by the Ammer River. In general, this clast-supported gravel shows no orientation or obvious gradation of clasts and no evidence of vegetation. When encountered, the bottom contact to the bedrock is sharp (Fig. 3g), while the gravel changes gradually upwards into a gray, loamy clay with few very coarse gravel clasts.

The overlying “Lower Clay” lithofacies ranged in thickness from 0.5 to 3 m, with an average depth of 9 m below ground surface. The dense, plastic clay was moist in the cores and very hard after drying. The top lithological contacts were partially gradational and partially sharp with indication of paleo-exposure in few cores.

The “Tufa” lithofacies is the thickest (2–9 m below ground surface) and most heterogeneous layer in the sedimentary floodplain sequence, and comprises silt to coarse-gravel sized calcareous grains (0.06–30 mm), which are referred to here as Tufa, with high content of organic matter, 5–50% fraction of total organic carbon (Fig. 3c), and imbedded peat layers (Fig. 3d). Three subfacies are distinguished based on color, grain size, and the presence of peat. In general, the upper Tufa sequence is fine-grained with very white color and millimeter-sized black organic-carbon specs. The middle section of the Tufa sequence often comprises brown coarser-grained calcareous deposits with grain sizes of 0.2–0.6 cm with hollow cylindrical nodules and peat inclusions up to 4 cm as well as abundant plant remnants, and planaspiral and trochospiral gastropodal shells. The Tufa middle section is characteristically cyclic, with each cycle (tens of centimeters) showing gradational changes in color (cream to light brown to dark brown), typically ending with a 5–10-cm-thick peat layer sharply contacting the next cycle. The lowermost section of the Tufa facies was often found to be dominated by dark brown, medium to coarse calcareous sands in a silty matrix. Thicker peat layers are predominantly found in the lower half of the Tufa succession, ranging in thickness from a few centimeters to a meter, with fully preserved and horizontally oriented leaves and wood fragments.

The top 2 m in the floodplain sequence consist of a very loamy, gray to brown clay with color transitions from lighter to darker brown with depth (Fig. 3c), which are interpreted here as alluvium. In this “Upper Clay” lithofacies, no shells or vegetation were observed and the moisture content was well preserved in the core. In the vicinity of the recent Ammer River course, the Upper Clay is thicker (up to 4 m), with evidence of well-rounded medium-sized gravels.

Along the southern fringes of the floodplain a “Hillslope Gravel” lithofacies is present, ranging in thickness from 4 to 13 m, and described as reddish-green clayey gravels with mudstone and sandstone clasts. The oligomictic gravel clasts were mostly edgy and up to 5 cm in size, indicating a shorter transport distance for the clasts. This poorly sorted lithofacies is highly variable between drilling locations but consistently showed no evidence of limestone clasts, differentiating from the Gravel floodplain lithofacies. Therefore, the authors interpret these clasts as originating from the bedrock of the hillslopes. At locations along the fringes of the floodplain, the hillslopes lithofacies are found in lithological contact with all central floodplain lithofacies.

The ability to detect geological features and their horizontal continuity by drill cores is limited by their spacing. To fill this gap, the previously presented geological datasets were complemented by geoelectrical surveying. The electrical resistivity tomography (ERT) profile across the floodplain shown in Fig. 4a confirmed the generally continuous layering of the floodplain sediments in the upper 10 m. Below, a relatively higher resistive anomaly of approximately 150 m width in the center of the floodplain indicates lateral changes in the basal gravel. The dashed white lines in Fig. 4 highlight the suspected outline of the basal-gravel feature, which requires ground-truthing by drill cores for validation. Subsequent drilling into the center of the anomaly revealed a clean gravel section in the lower 5 m above a 20 m deep bedrock contact. This higher-resistivity feature could be traced over the distance of about 1 km by depth-targeted geoelectrical mapping with constant electrode spacing (Klingler et al. 2020). Figure 4b presents the results of the geoelectrical mapping with red colors representing higher apparent resistivities in the target depth. This meandering belt of higher apparent resistivities is interpreted as a paleo-channel structure of cleaner gravel bounded by lower-resistive mudstones of the bedrock (in blue in Fig. 4b).

Geological modelling facilitated the visualization of the spatial extent, continuity and thicknesses of the floodplain stratigraphy, incorporating all geological and geophysical data, along with surficial geological maps (LGRB 2012), and a 10-m digital elevation model (LGL 2009). Figure 5a presents “along” and “cross-valley” cross-sections, exported from the geological model. Figure 5b shows the profile-lines of the cross-sections together with the drilling locations used for the construction of the geological model. Both cross-sections in Fig. 5a display the extensive horizontal layering of the central floodplain lithology, with the highest variations in thickness towards the hillslopes. In the cross-valley direction, the Tufa and Gravel facies generally thicken towards the middle of the floodplain, whereas the Lower Clay becomes thinner. The Upper Clay is relatively uniform in thickness; however, a thickening trend is evident near the Ammer River, as observed from the drill cores. Moving towards the southern hillslope, the floodplain lithofacies pinch out as supported by vertical logs of natural gamma radiation marked in the cross sections (Fig. 5a). The across-valley cross-section is in line with a small tributary valley. Here, the lithofacies of the tributary-valley filling, denoted “Hillslope Aquifer” in the legend of Fig. 5, fingers into the floodplain and seems to be in contact with all floodplain facies.

Based on the lithological (primary and secondary textures) and geometric parameters (extent and continuity), the geological features were grouped into hydrostratigraphic units. The uppermost fine-textured alluvial deposit is considered a regional aquitard (Upper Clay Aquitard; Fig. 5c). Due to capillary forces, the lower half of the alluvium remains water-saturated for most of the year, whereas the top meter shows shrinkage cracks in dry summer/fall periods. The latter may facilitate temporary preferential flow paths during summer storm events.

The underlying tufa sediments are highly heterogeneous and permanently water-saturated. As the horizontal continuity of individual layers is uncertain, the entire Tufa sequence was interpreted as a single regional aquifer (“Tufa Aquifer”; Fig. 5c). Monitoring wells were installed to target the coarse-grained sections in the upper 2–3 m of the Tufa Aquifer. Well tests revealed estimates of aquifer transmissivity covering four orders of magnitude; however, the majority fell within the range of 10⁻⁴ to 10⁻⁶ m²/s, with a geometric mean on the order of 1.8 × 10⁻⁵ m²/s (Fig. 5c).

The Tufa Aquifer is continuously underlain by the Lower Clay, which is interpreted as a regional aquitard (Lower Clay aquitard; Fig. 5c). It separates the tufa sediments from the lowermost Gravel lithofacies, which extends across the majority of the floodplain and is interpreted as a regional aquifer (“Gravel Aquifer”; Fig. 5c). A second set of monitoring wells targeting the Gravel Aquifer was installed over its full thickness. Though Gravel Aquifer transmissivity estimates from well tests span three orders of magnitude, the vast majority are near 10⁻⁴ m²/s, resulting in a geometric mean on the order of 1.3 × 10⁻⁴ m²/s (Fig. 5c). These values are quite small for a gravel body but consistent with the high clay content. In the clean gravel channel at depth (marked in dark yellow in Fig. 5), determined by geoelectrical monitoring, the aquifer transmissivity was substantially higher, ≈ 1.2 × 10⁻³m²/s.

The hillslope lithofacies within the southern tributary valleys are extremely heterogeneous, containing clayey material to gravel-sized clasts. Coarser material, found at the basal interface with the bedrock, are water-saturated and considered as an aquifer (Hillslope Aquifer; Fig. 5c). Monitoring wells targeting shallow and deep hillslope water-bearing sediments were installed and well tests were performed, yielding highly variable estimates of aquifer transmissivity ranging from 1 × 10⁻⁷ to 2.5 × 10⁻⁴ m²/s (Fig. 5c).

Figure 6 presents the groundwater contour maps from July 2019 for the Tufa and Gravel aquifers, constructed from observations in 31 and 21 monitoring wells, respectively. The dashed contour lines refer to estimated groundwater levels in the hillslopes, which were constructed by including the hillslope monitoring wells in the kriging estimate of groundwater levels.

Groundwater flow in the Tufa Aquifer is mainly oriented along the valley (along-valley, Fig. 6a). Close to the southern boundary and within the hillslopes, a significant cross-valley component exists, suggesting that shallow groundwater from the southern hillslope enters the Tufa Aquifer, is translated towards the center of the floodplain, and then subsequently transferred down-valley. At the northern boundary, a strong curvature in the groundwater elevation contours is most evident in the eastern portion of the Tufa Aquifer near the Himbach Valley (eastern end in Fig. 6a), indicating an influx of groundwater from this tributary valley. Groundwater flow within the Gravel Aquifer is predominantly along-valley (Fig. 6b), while similar to the Tufa Aquifer, there is some cross-valley groundwater flow in the eastern portion of the aquifer, whereas the influence of the southern boundary appears to be smaller in the Gravel Aquifer than in the Tufa Aquifer.

As shown in Fig. 2, the southern hillslope extends all along the Wurmlingen Saddle and is mainly covered by agricultural land. Water infiltrating along this hillslope most likely contains solutes of agricultural origin that may enter the floodplain groundwater, having a bigger impact on flow in the Tufa than in the Gravel Aquifer, as indicated by the groundwater contour maps.

The origin of the substantial groundwater discharge, entering the study area at its western upstream end into both floodplain aquifers, is beyond the scope of the current study. This water may have been recharged by the Ammer River at the location where the floodplain valley widens, it may be contributed by inflows from the karstic Muschelkalk aquifer at the fault line west of the study area, or it may originate from the northern tributary valley of the Käsbach stream. Local groundwater recharge in the floodplain itself is limited by the extended coverage by alluvial fines in the entire floodplain.

Figure 7 compares weekly averaged data of groundwater level, precipitation, and river stage to reveal the spatiotemporal dynamics of the floodplain and hillslope groundwater systems. Until June 2019, the seasonal dynamics in both groundwater systems are characterized by four main time periods, denoted a–d in Fig. 7, whereby periods a and c are characterized by declining groundwater levels, while periods b and d show rising groundwater levels. The periods reflect seasonal fluctuations in precipitation and evapotranspiration, causing little groundwater recharge from late spring until fall, and enhanced groundwater recharge in the winter months. Note that the winter 2017/2018 was exceptionally wet and the year 2018 extremely dry from late spring until fall. As a consequence, the high-water-table period b ended in March 2018, whereas the similar period d extended until the end of the period discussed here (June 2019), which was caused by strong precipitation events in the late spring of 2019. Figure 7a illustrates that the shallow and deep monitoring wells in the hillslope (X016) and floodplain (X019, X015 and X012) aquifers differ in their temporal dynamics, with the shallow wells responding more quickly to precipitation events in the wet periods b and d. Among all monitoring wells, the shallow hillslope wells showed the strongest response to precipitation events in times of high water tables.

Vertical hydraulic gradients between the Tufa and Gravel aquifers vary both in space and over the seasons. In the up-valley floodplain (X019) and the hillslope (X016) locations, the vertical hydraulic-head differences are generally small, except in response to precipitation events in the high-level periods b and d, when water tables in the shallow observation wells are higher than in the deep wells. In the two down-valley floodplain locations (X015 and X012), the hydraulic heads are significantly higher in the deep wells compared to the shallow ones in the periods a and c of declining groundwater levels, whereas the differences are much smaller during the wet periods b and d.

Altogether, Fig. 7 shows that groundwater levels in the hillslope and the floodplain exhibit essentially the same seasonal trends, whereas fluctuations of the river hydrograph and the groundwater levels hardly correlate. The hillslope groundwater levels show the strongest and fastest response to individual rainfall events, followed by the Tufa Aquifer, and finally the Gravel Aquifer, for which responsiveness strongly depends on the season.

Figure 8a shows scatter plots of sulfate, bicarbonate, magnesium, calcium concentrations, and groundwater electrical conductivity (EC). Figure 8b contains a hydrochemical cluster analysis of all samples, whereas Fig. 8c,d present maps of EC in the two floodplain aquifers.

Floodplain and hillslope groundwater samples of the July 2019 campaign showed positive correlations between EC and calcium, magnesium and sulfate concentrations, whereas no clear relationship between EC and bicarbonate concentrations can be observed (Fig. 8a). EC clusters at approximately 1,200 μS/cm, and some samples reach up to 3,000 μS/cm. A near 1:1 relationship between sulfate and calcium, especially in samples with EC > 1,200 μS/cm, indicates gypsum dissolution (red box, Fig. 8a), which can be explained by the gypsum-bearing mudstones of the Grabfeld formation.

Figure 8b presents results of a hierarchical cluster analysis based on an extended hydrochemical dataset, including more hydrochemical parameters than shown in the scatter plots of Fig. 8a. All parameters were z-score normalized prior to the cluster analysis. Three clusters at a statistical distance of 10 were identified (orange line in Fig. 8b). In the maps of Fig. 8c,d, the sampling locations belonging to different clusters are visualized by different marker symbols.

Cluster I comprises highly mineralized (predominantly calcium and sulfate) groundwater samples (black circles, Fig. 8a) and the sample from the northern spring fed by the Grabfeld formation (red circle in Fig. 8a and red bar in Fig. 8b). This cluster includes samples of both the Tufa and Gravel aquifers in northern portions of the floodplain, specifically at the confluencing tributary valleys, where EC is particularly high (Fig. 8c,d). These samples are strongly influenced by gypsum dissolution.

Cluster II consists of low mineralized groundwater samples (black crosses, Fig. 8a) and a sample of the spring at the southern hillslope (red cross in Fig. 8a and red bar in Fig. 8b). These samples originate from the southern Hillslope Aquifers or from southern-most monitoring wells in the floodplain aquifers, confirming the interpretation of the groundwater contour maps (Fig. 6) that the floodplain aquifers receive water from the southern hillslopes. Due to the correlation in oxidized compounds, this cluster also contains surface water samples collected from the Ammer River and the Mühlbach stream (blue circles in Fig. 8a and blue bars in Fig. 8b); however, these surface-water samples are far more similar to the central floodplain in major ion chemistry (Fig. 8a). Cluster III (Fig. 8b) contains the largest number of samples, mainly from the central floodplain (black triangles, Fig. 8a), and includes samples of both the Tufa and Gravel aquifers (Fig. 8c,d).

Figure 9 shows scatterplots of redox-indicating concentrations versus field-measured redox potentials (Eh) of the sampling campaign in July 2019 as well as maps of interpolated Eh values in the Tufa and Gravel aquifers. Like in Fig. 8, the Eh-map of the Tufa Aquifer in Fig. 9a includes measurements of shallow observation wells in the Hillslope Aquifer, and the Eh-map of the Gravel Aquifer in Fig. 9b includes those of deep observation wells in the Hillslope Aquifer. The marker symbols in the scatter plots and the maps refer to the hydrochemical cluster affiliations of the samples discussed previously.

The scatterplots of Fig. 9a,b confirm that field-measurements of Eh yield a good qualitative assessment of redox-sensitive species, in which high Eh values indicate low concentrations of the reduced species bisulfide and ammonium and high concentrations of electron acceptors nitrate and dissolved oxygen. Dissolved organic carbon (DOC) showed a much weaker relationship with Eh. In many scatter plots, the previously described clustering is quite evident, especially with groundwater samples from the southern portion of the floodplain and the Hillslope aquifer (black crosses, Fig. 9a,b). A quantitative analysis of defined redox pairs (bisulfide/sulfate and ammonium/nitrate) by the Nernst equation, however, yielded calculated redox potentials that were substantially lower than the field measurements so that the field-measured Eh values cannot be taken to compute the electron buffering capacity of the solutions.

In the maps of interpolated Eh values in Fig. 9, blue regions indicate positive field-measured Eh values and red regions negative ones. The Eh map of the Tufa Aquifer in Fig. 9a shows oxidizing conditions in the hillslope samples, with mean Eh, DO and nitrate concentration of 83 mV, 3.2 mg/L, and 13 mg/L, respectively. The width of a transition zone with field-measured Eh values around zero, marked white in the map, is small (≈ 100 m) and uncertain because the underlying interpolation by kriging is smooth and the network of monitoring wells was not optimized to resolve the redox transition zone. Practically the entire Tufa Aquifer shows strongly reducing conditions. In the center were observed field-measured Eh-values as low as −102 mV, and ammonium and bisulfide concentrations as high as 19 and 47 mg/L, respectively. This is consistent with the Tufa containing large amounts of labile organic carbon acting as strong electron donor in the matrix. Two samples north of the Ammer River, and one south (eastern edge) show less reducing conditions, which may indicate an influence of the river.

Figure 9b shows the interpolated Eh map of the Gravel Aquifer, which differs from that of the Tufa Aquifer. Groundwater in the middle section of the Gravel Aquifer is again strongly reduced (average bisulfide concentrations: 15 mg/L, field-measured Eh as low as −83 mV). Hillslope samples are again oxidized, but also the sample at the western inflow of the study domain and samples at the downstream eastern end. Here, the mean values of Eh, DO and nitrate concentrations were 66 mV, 1.9 mg/L and 18 mg/L, respectively. The sediments in the Gravel Aquifer contain less organic matter than the Tufa, explaining a longer transition zone from oxidizing to reducing conditions, even though the true transition at the western end is not well resolved. The transition from reduced to oxidized conditions in the direction of groundwater flow from the west to the east is more difficult to explain as there cannot be an internal source of dissolved oxygen. It is believed that the channel of clean gravel underneath the clayey gravel, detected by the geoelectrical mapping campaign, acts as a preferential pathway transferring more oxidized water underneath the main sediment body with little or no chemical reduction. This channel has not been met by most wells in the center of the Gravel Aquifer, except for X072 (marked with a star in Fig. 9b). Further downstream, all wells in the Gravel Aquifer are oxidized, carrying the signature of the water in the clean gravel channel. This implies that the relative contribution of the groundwater flux through the reduced parts of the Gravel Aquifer must be small. Without such a bypassing mechanism, the observed redox pattern cannot be explained.

The floodplain is carved into the gypsum-bearing Triassic mudstone of the Grabfeld formation. This relatively incompetent bedrock is susceptible to weathering and facilitated the development of the wide and thick floodplain sequence observed at the study site in present day. Up- and downstream of the study area, the bedrock is much more competent (Muschelkalk limestone and sandstone of the Stuttgart Formation, respectively). Correspondingly, the floodplain becomes narrower and shallower. This characteristic landscape of alternating narrow and wide floodplains, controlled by the changing lithology along a river course, is quite common in European upland catchments with clastic bedrock. At the upstream ends of the wider basins, river water tends to infiltrate, whereas groundwater discharges back to the river at the downstream ends, triggering basin-scale riparian exchange (Fig. 10a).

The southern hillslopes extend over a length of up to ≈ 1 km (Wurmlingen Saddle, Fig. 2), whereas the northern hillslopes are comparably short and steep. Along the Wurmlingen Saddle, a substantial weathering zone of up to 20 m thickness has been confirmed by drill cores, which is common in landscapes with hilly topography and bedrocks of low competence (Rempe and Dietrich 2014). The southern hillslopes are also characterized by pronounced hillslope hollows, which consist of thick, fine-grained, poorly sorted colluvium, produced by solifluction processes typical of the periglacial environment (Collins et al. 2006).

On the steep northern slopes, the bedrock is less weathered and correspondingly less permeable. Two large tributary valleys confluence the floodplain at the upstream and downstream end of the study area. Here the hillslope deposits are interpreted as alluvial fans and thus expected to contain coarser-grained sediments in comparison to the southern slopes. Altogether, hillslope deposits, particularly those in hillslope hollows or tributary valleys, can collect regional groundwater recharge along the hillslopes and transfer them into both floodplain aquifers as these deposits pinch out over the entire thickness of the floodplain filling.

The location of the groundwater divide on the southern hillslope is unclear and depends on the exact thickness of the weathering layer. In principle, the Neckar Valley south of the Wurmlingen Saddle is topographically lower, which may shift the groundwater divide into the Ammer Valley, provided that sufficiently permeable material reaches deep enough (Kortunov 2018). Such ambiguities are fairly typical for mudstones in hilly topographies and have a large influence on the local water balance.

The fluvial systems of many European rivers have shown significant response to the environmental change from the late Pleistocene to Holocene time periods, which is captured in the fluvial sediment record (Collins et al. 2006; Lespez et al. 2008). Gibbard and Lewin (2002) present a conceptual geological model for floodplain stratigraphy common in small- to medium-sized European rivers: high-energy late Pleistocene gravels preserved at the base, followed by the deposition of fine-grained, organic-rich (and possibly tufaceous) sediments from the warmer Atlantic period in the Holocene, and finally a thick anthropogenically impacted alluvium blanketing the present-day floodplain. The subsequent interpretation, of the collected geological data, supports this representative floodplain stratigraphy (from oldest to youngest):

While the general depositional sequence of the Ammer floodplain is fairly common, its local specification determines the hydrogeological functioning of the floodplain (right side, Fig. 1). This particularly refers to the extent and connectivity of highly permeable clean gravels (flowpath D, Fig. 1), their connection to hillslope deposits (flowpath A, Fig. 1), and the intactness of the alluvial fines in the vicinity of the river (flowpath B and C, Fig. 1). If the coarser-grained deposits of the active river do not cut through the alluvial fines, the river remains decoupled from the groundwater. Also, clean-gravel deposits act as the “drainage pipe” of the floodplain, making a substantial fraction of the water flux bypass the reducing floodplain sediments.

As illustrated in Fig. 10b, water balances for the two floodplain aquifers under steady-state conditions were set up, resulting in the following balance equation:in which Q_g,CPhslp is the groundwater flux entering at the control plane at the southern hillslope-floodplain boundary, Q_g,CPupstr is the groundwater flux entering at the western boundary of the study domain, and Q_g,CPdwnstr is the groundwater flux leaving at the eastern boundary of the study domain. The location of the upstream and downstream boundaries makes the groundwater inflows from the northern tributary subcatchments (Käsbach and Himbach, see Fig. 2) irrelevant to the presented floodplain water balance. Furthermore, this study did not consider any groundwater recharge within the floodplain itself because the alluvial fines are hardly permeable.

The groundwater fluxes across control planes, Q_g,CPi, are evaluated by integrating the specific discharge normal to the plane over the surface area of the plane:in which w_CPi is the width of control plane i, n is the unit normal vector, ∇h is the hydraulic gradient estimated from the kriging interpolation, K is the hydraulic conductivity derived from single-well pumping tests, and m is the thickness of the aquifer derived from the geological model.

Table 2 lists results of the steady-state water balance, with long-term averages, peak groundwater flows and solute fluxes of total nitrogen and sulfur. For determining long-term averages of groundwater fluxes, monthly groundwater contour maps were generated for both floodplain aquifers, from May 2018 until July 2019. Two scenarios were considered—in scenario 1, the geometric mean of transmissivity estimates in the Tufa and Gravel aquifers, and an arithmetic mean for the Hillslope Aquifers, were taken as the basis for the aquifer transmissivity estimates; however, in scenario 2, higher estimates were considered. For the Hillslope Aquifer, transmissivities twice that of the arithmetic mean were assigned to aquifer material within the thick hillslope hollow colluvium. These values are comparable to the highest yielding hillslope monitoring wells. The transmissivity of the Tufa Aquifer was also increased by about a factor of four, and the study explicitly accounted for the channel of clean gravel, where transmissivity values are determined from two monitoring wells that are potentially screened in this channel (1.2 × 10⁻³ m²/s rather than the geometric mean of 1.3 × 10⁻⁴ m²/s of the entire Gravel Aquifer).

With an annual groundwater recharge of approximately 200 mm/year, and a recharge area of 2.65 km², the floodplain collects less than 1% (~0.12 L/s) of the total hillslope recharge (16.8 L/s) in scenario 1, and approximately 2–4% in scenario 2. Thus, this groundwater flux is collected within the colluvial aquifer material in the hillslope hollows and subsequently partitioned between the floodplain aquifers. Approximately 15 and 85% of the hillslope groundwater goes to the Tufa and Gravel aquifers, respectively. In both scenarios, it is estimated that the hillslope infiltrate increases the along-valley groundwater discharge within the Tufa and Gravel aquifers by approximately 100–150%, from CP_upstr to CP_dwnstr. In scenario 1, groundwater discharge in the Gravel Aquifer is about three to four times larger than that in the Tufa Aquifer, whereas this increases to a factor of six in scenario 2 when accounting for the gravel channel.

Under these conceptual assumptions, the water balance calculations suggest that the recharge area contributing to the local floodplain groundwater system is very small in comparison to the entire southern hillslope between CP_upstr to CP_dwnstr, and the majority of hillslope infiltrate completely bypasses the floodplain groundwater system. It is therefore concluded that the bulk hydraulic properties associated to the predominantly fine-grained hillslope and floodplain sediments do not facilitate enough groundwater flow to drain the southern hillslope recharge area and the floodplain would rather act as a barrier to cross-valley groundwater flow than as a conduit.

As seen in the comparison between scenarios 1 and 2, the groundwater fluxes are highly sensitive to the geometric and hydraulic properties of the aquifer materials. The uncertainty of such properties is extremely high in these characteristically heterogeneous geologic settings in temperate climates. Increasing the effective hydraulic conductivity and connectivity of the hillslope hollow colluvium and the downstream gravel channel by a factor of three to four would result in far different interpretations on the hydraulic behavior of the floodplain, thus its biogeochemical functioning.

Peak groundwater discharge estimates presented in Table 2 highlight the increased discharge dynamics within the Hillslope Aquifer as compared to the floodplain aquifers. Groundwater discharge from the hillslope increased as much as 0.37 L/s as compared to ~0.05 L/s in the floodplain. These hillslope dynamics are observed both in seasonal groundwater fluctuations and from single rainfall events (Fig. 7). Since the floodplain aquifers do not accommodate the extra flux, the springs and drainage channels running parallel to the hillslope-floodplain boundary (Fig. 2) must act as “release valves” and receive a substantial fraction of subsurface hillslope runoff during hydrological events.

Table 3 presents surface water and groundwater fluxes from a rainfall event on March 13, 2019. Over the course of a day and a half, there was 10 mm of rainfall, followed by a 3-week dry period. Surface-water fluxes are derived from river stage measurements in v-notch weirs within the drainage channels, as outlined in Fig. 10b, while groundwater fluxes are calculated from the groundwater contour maps. Pre- and post-rainfall water fluxes show the increase in discharges immediately after rainfall, which are summed up over a subsequent 14-day period, to estimate the approximate amount of additional water coming from the rainfall event. The increase in discharge within the drainage channels is substantially higher than the groundwater flux increase, indicating that a significant amount of recharge from the rain event likely travels rapidly to the drainage channels as interflow. Increases in hillslope groundwater fluxes were twice as large as in the two floodplain aquifers combined, so it is believed that this additional groundwater flux is captured by the hillslope spring. Though the hillslope spring is providing an important release valve for excess hillslope groundwater, the total amount of water which accumulated over the days following the event, only accounts for approximately 15% (960 m³) of the total hillslope recharge (6,600 m³) from that rain event. The discharge estimates from the drainage channel weirs are uncertain. However due to the large discrepancy between groundwater and surface-water fluxes, as well as surface-water fluxes and hillslope recharge, this measurement uncertainty does not impact the interpretation of the results.

The likely candidate for this substantial unaccounted-for hillslope recharge could be a weathered bedrock aquifer, which may discharge further downstream to the gravel channel, or possibly to the neighboring river catchment as inter-basin groundwater flow.